thursday-agent 0.5.0 → 0.7.0

package/features/bot/bot.runner.ts

@@ -1,28 +1,16 @@
+import { setTimeout as wait } from "node:timers/promises";
 import type { ModelMessage } from "ai";
+import { appEvents, presence } from "@/app/api/events/app-event.server";
+import { BOT_RUN, WORKSPACE_KEEP } from "@/config";
+import { isProviderRefusal, modelErrorToString } from "@/features/ai/model";
+import { buildThreadOpening } from "@/features/ai/prompts/bot.prompt";
+import { isAnyCallLive } from "@/features/thursday/thursday.query";
+import { opensOnFinish, pathsIn } from "@/features/workspace/file-kind";
 import {
-  appEventStream,
-  appEvents,
-  presence,
-} from "@/app/api/events/app-event.server";
-import { BOT_RUN, PATHS, WORKSPACE_KEEP } from "@/config";
-import {
-  type ModelFailure,
-  modelErrorToString,
-  modelFailureOf,
-} from "@/features/ai/model";
-import {
-  buildTaskOpening,
-  OPENING_TURNS,
-} from "@/features/ai/prompts/bot.prompt";
-import { TOOL_NAMES } from "@/features/ai/tools/tool-name";
-import {
-  isAnyCallLive,
-  listCallTurns,
-} from "@/features/thursday/thursday.query";
-import { pathsIn } from "@/features/workspace/file-kind";
-import {
+  botBrowserSession,
   closeHiddenBrowser,
   closeJobShell,
+  filesOnDisk,
   jobScratch,
   listScratchFolders,
   pruneJobFiles,
@@ -31,302 +19,469 @@ import {
 } from "@/features/workspace/workspace";
 import { desktopNotify } from "@/lib/desktop-notify";
 import { logger } from "@/lib/logger";
-import { publicError } from "@/lib/public-error";
+import { isPublicError, publicError } from "@/lib/public-error";
 import { createKeyedLock } from "@/lib/queue";
+import { PromiseChain } from "@/lib/utils";
+import { findJobBot } from "./bot.query";
+import { resumeTranscript, runBot, type ThreadEvent } from "./bot.run";
 import {
-  createBotDesks,
-  resumeThread,
-  runBot,
-  type TaskEvent,
-} from "./bot.run";
-import {
-  TASK_CONTINUE,
-  type TaskPending,
-  type TaskSpeaker,
-  type TaskStatus,
-  tagSpeaker,
+  isAppStop,
+  THREAD_CONTINUE,
+  type ThreadSpeaker,
+  type ThreadStatus,
 } from "./bot.schema";
 import {
-  addTaskUsage,
-  countStopsSinceSpoken,
+  appendRoomMessage,
+  cancelRoom,
+  claimRoomWork,
+  consumeRoomInbox,
+  finishRoomWork,
+  listParticipantTranscript,
+  listRoomReceipts,
+  listRoomWork,
+  lowerRoomContextBudget,
+  noteRoomBreak,
+  pauseRoom,
+  type RoomWork,
+  resumeRoom,
+  roomContextBudget,
+  sendRoomMessage,
+  settleRoom,
+  tellRoom,
+} from "./room.query";
+import { ROOM_THURSDAY } from "./room.schema";
+import {
+  addThreadUsage,
   deleteMessages,
-  deleteTask,
-  findTask,
-  insertTask,
-  lastSeq,
-  listAutoStoppedTasks,
-  listBorrowedTails,
-  listRunningTaskIds,
-  listTaskFolders,
-  listThread,
-  stopNote,
-  type TaskMessageInput,
-  updateTask,
+  deleteThread,
+  findThread,
+  insertThread,
+  listAutoStoppedThreads,
+  listRunningThreadIds,
+  listThreadFolders,
+  updateThread,
   upsertMessage,
-} from "./task.query";
-
-/**
- * Runs a job so it outlives the request, writing every event to the task's
- * thread (task_message). A run always starts from the thread. `answerTask` is
- * the one way a person adds to it, serialized per job by `taskLock`, and
- * `resumeTask` the one way the app picks a job back up by itself.
- */
-
-type Run = {
-  stop: AbortController;
-  /** Resolves when the run ends, however it ends. Interrupters wait on it. */
-  done: Promise<void>;
-  /**
-   * Queues words for a run still going (bot.run `notes`). False once the loop
-   * has ended; then they go on the thread the ordinary way (answerTask).
-   */
-  note: (text: string) => boolean;
-};
+} from "./thread.query";
 
+type Run = { threadId: string; stop: AbortController; done: Promise<void> };
 type Pinned = {
-  __botRuns?: Map<string, Run>;
-  __botTaskLock?: ReturnType<typeof createKeyedLock>;
-  __botResumes?: Map<string, ReturnType<typeof setTimeout>>;
-  __botPausing?: { current: Promise<void> };
+  __roomRuns?: Map<string, Run>;
+  __roomThreadLock?: ReturnType<typeof createKeyedLock>;
+  __roomPausing?: { current: Promise<void> };
 };
-
-/** Pinned to globalThis: next dev reloads this module under a running run, and the abort controller and lock must be the originals. */
-const running: Map<string, Run> = ((globalThis as Pinned).__botRuns ??=
-  new Map());
-
-/**
- * One lane per job. Stop, append, relaunch must not interleave: two answers a
- * moment apart would abort the same run and claim the same `lastSeq + 1`.
- */
-const taskLock = ((globalThis as Pinned).__botTaskLock ??= createKeyedLock());
-
-/** When each parked job picks itself back up (scheduleResume); one timer per job. */
-const resumes: Map<string, ReturnType<typeof setTimeout>> = ((
-  globalThis as Pinned
-).__botResumes ??= new Map());
-
-/** The pause under way, if any; a returning browser waits for it (resumeStoppedTasks). */
-const pausing = ((globalThis as Pinned).__botPausing ??= {
+const running = ((globalThis as Pinned).__roomRuns ??= new Map<string, Run>());
+const threadLock = ((globalThis as Pinned).__roomThreadLock ??=
+  createKeyedLock());
+const pausing = ((globalThis as Pinned).__roomPausing ??= {
   current: Promise.resolve(),
 });
 
-/** Opens a job. The opening message — who is on it, the job, the call it came from (bot.prompt buildTaskOpening) — is the thread's first row. */
-export async function startTask(input: {
+export async function startThread(input: {
   bot: string;
   request: string;
   label: string;
   callId?: string | null;
-  /** Who handed it over: Thursday during a call, or the user on screen. */
-  from: TaskSpeaker;
+  from: ThreadSpeaker;
 }) {
-  const { from, ...row } = input;
-  const conversation = row.callId
-    ? await listCallTurns(row.callId, OPENING_TURNS)
-    : [];
-  const opening = buildTaskOpening({
+  const found = await findJobBot(input.bot);
+  if (!found || found.disabled) publicError("Choose an enabled bot.");
+  const { from, ...row } = { ...input, bot: found.name };
+  const opening = buildThreadOpening({
     bot: row.bot,
     request: row.request,
-    conversation,
     from,
   });
-  const task = await insertTask({ ...row, opening });
-  launch(task.id, {
-    bot: row.bot,
-    messages: [{ role: "user", content: opening }],
-  });
-  return task.id;
+  const thread = await insertThread({ ...row, opening });
+  await pump(thread.id);
+  return thread.id;
 }
 
-/**
- * A person saying something to a job, one path for three cases: an answer to
- * a job stopped on `ask_thursday`, a note to a running one (queued and read
- * before its next step; the step in flight finishes), or a follow-up to a
- * finished one. Only when the run has already ended does the running case
- * fall through to stop, append, relaunch.
- *
- * @param from Who said it. The words go on the thread tagged with it (bot.schema
- * tagSpeaker): Thursday passing something on from a call and the user typing on
- * screen reach the bot down this one pipe, and a go-ahead means something
- * different from each. Unset for words that already carry their tags — the notes
- * a finished run left unread (drive).
- */
-export async function answerTask(
+/** A recipient selects a desk; replying to a question also names the exact exchange. */
+export async function answerThread(
   id: string,
   answer: string,
-  from?: TaskSpeaker,
-) {
-  const said = from ? tagSpeaker(from, answer) : answer;
-  return taskLock(id, async () => {
-    const task = await findTask(id);
-    if (!task) publicError("No such job.");
+  from: ThreadSpeaker = "user",
+  recipient?: string,
+  replyTo?: string,
+): Promise<Awaited<ReturnType<typeof tellRoom>> | null> {
+  let told: Awaited<ReturnType<typeof tellRoom>> | null = null;
+  await threadLock(id, async () => {
+    const thread = await findThread(id);
+    if (!thread) publicError("No such thread.");
+    if (recipient) {
+      const participants = await listRoomWork(id);
+      const found = participants.find(
+        (row) => row.bot.toLowerCase() === recipient!.trim().toLowerCase(),
+      );
+      if (!found || found.bot === ROOM_THURSDAY)
+        publicError("Choose a participant in this thread.");
+      recipient = found.bot;
+    }
+    // Continue picks up a stop the app made; to a bot's question it is an answer like any other
+    if (answer.trim() === THREAD_CONTINUE && isAppStop(thread.pending)) {
+      await resumeRoom(id);
+      const all = await listRoomWork(id);
+      if (!all.some((row) => row.state === "queued"))
+        told = await tellRoom(
+          id,
+          "Continue from the saved conversation.",
+          from === "user" ? "The user" : ROOM_THURSDAY,
+          recipient,
+          undefined,
+          false,
+        );
+    } else {
+      told = await tellRoom(
+        id,
+        answer,
+        from === "user" ? "The user" : ROOM_THURSDAY,
+        recipient,
+        replyTo,
+      );
+    }
+  });
+  await pump(id);
+  return told;
+}
 
-    const live = running.get(id);
-    if (live && task.status === "running" && live.note(said)) {
-      logger.debug(`task ${id}: interjection queued`);
+/** Claiming is short and serialized; model execution never holds the room lock. */
+async function pump(id: string) {
+  await threadLock(id, async () => {
+    if (!presence.watching) {
+      const thread = await findThread(id);
+      if (
+        thread?.status === "running" &&
+        ![...running.values()].some((run) => run.threadId === id)
+      )
+        await pauseRoom(
+          id,
+          "The app is not open. Work resumes when it returns.",
+          true,
+        );
       return;
     }
-    if (live) {
-      live.stop.abort();
-      await live.done;
+    for (let pass = 0; pass < 2; pass++) {
+      let work: RoomWork | null;
+      while ((work = await claimRoomWork(id))) launch(work);
+      await settleRoom(id);
     }
-    // A person picked it up; whatever the app meant to do by itself is moot
-    forgetResume(id);
-
-    const thread = await listThread(id);
-    const { reply, as } = joinThread(task, said, thread);
-    logger.debug(`task ${id}: ${as}`);
-
-    await upsertMessage(id, (await lastSeq(id)) + 1, {
-      bot: null,
-      parent: null,
-      ...reply,
-    });
+  });
+}
 
-    await updateTask(id, {
-      status: "running",
-      outcome: null,
-      pending: null,
-      seen: false,
-      endedAt: null,
-    });
-    // Resumed on the rows as they were stored (bot.run resumeThread)
-    launch(id, {
-      bot: task.bot,
-      messages: resumeThread([...thread, reply], await listBorrowedTails(id)),
-    });
+function launch(work: RoomWork) {
+  const stop = new AbortController();
+  let finish!: () => void;
+  const done = new Promise<void>((resolve) => {
+    finish = resolve;
   });
+  running.set(work.id, { threadId: work.threadId, stop, done });
+  void drive(work, stop.signal)
+    .catch((cause) =>
+      logger.error(`thread ${work.threadId}: participant`, cause),
+    )
+    .finally(() => {
+      running.delete(work.id);
+      finish();
+      void pump(work.threadId).catch((cause) =>
+        logger.error(`thread ${work.threadId}: scheduling`, cause),
+      );
+    });
 }
 
-/**
- * The `ask_thursday` calls still open in the step a job stopped on: a model can
- * ask twice at once, and the one answer goes to each rather than leaving the
- * second to read as a call cut off (bot.run resumeThread). A call that already
- * has a result — arguments the tool refused — is not open, and the words go in
- * as a turn instead: a second result for one call is refused outright.
- */
-function openAsks(thread: ModelMessage[], stoppedOn: string): string[] {
-  const answered = new Set(
-    thread.flatMap((message) =>
-      message.role === "tool"
-        ? message.content.flatMap((part) =>
-            part.type === "tool-result" ? [part.toolCallId] : [],
-          )
-        : [],
-    ),
-  );
-  const step = thread.findLast(
-    (message) =>
-      message.role === "assistant" &&
-      typeof message.content !== "string" &&
-      message.content.some(
-        (part) => part.type === "tool-call" && part.toolCallId === stoppedOn,
-      ),
-  );
-  const asks =
-    step?.role === "assistant" && typeof step.content !== "string"
-      ? step.content.flatMap((part) =>
-          part.type === "tool-call" && part.toolName === TOOL_NAMES.ask_thursday
-            ? [part.toolCallId]
-            : [],
-        )
-      : [stoppedOn];
-  return asks.filter((toolCallId) => !answered.has(toolCallId));
+async function drive(work: RoomWork, signal: AbortSignal) {
+  let turn = await attempt(work, signal);
+  // A break gets one more try from the stored transcript; a refusal, the content
+  // filter and the step limit wait for a person, and so does a second break
+  if (turn?.failure?.retry && !signal.aborted) {
+    await noteRoomBreak(work, turn.failure.message);
+    // Only an abort rejects the wait, and the check below reads it
+    await wait(BOT_RUN.retryMs, undefined, { signal }).catch(() => {});
+    if (!signal.aborted) turn = await attempt(work, signal);
+  }
+  if (!turn || signal.aborted) return;
+  const failure = turn.failure?.message;
+  const final = turn.ending;
+  if (failure || !final || final.stopped) {
+    // Release this run before the room lock drains it; cancellation holds the same lock.
+    const why =
+      failure ??
+      "The turn was interrupted. Continue from the saved conversation.";
+    void threadLock(work.threadId, async () => {
+      const current = (await listRoomWork(work.threadId)).find(
+        (row) => row.id === work.id,
+      );
+      if (
+        current?.state !== "running" ||
+        current.generation !== work.generation
+      )
+        return;
+      const drained = stopRuns(work.threadId);
+      await pauseRoom(work.threadId, why);
+      await drained;
+    }).catch((cause) =>
+      logger.error(`thread ${work.threadId}: pausing`, cause),
+    );
+    return;
+  }
+  await finishRoomWork(work, final.text);
+  const thread = await findThread(work.threadId);
+  if (thread?.status === "done") {
+    if (!(await isAnyCallLive()))
+      desktopNotify(thread.label, thread.outcome ?? "");
+    const path = (await filesOnDisk(pathsIn(thread.outcome ?? ""), null)).find(
+      opensOnFinish,
+    );
+    if (path)
+      appEvents.emit({
+        type: "artifact",
+        threadId: thread.id,
+        label: thread.label,
+        path,
+      });
+  }
 }
 
-/**
- * How a person's words enter the thread. A job stopped on `ask_thursday` gets
- * them as that call's tool result (a call with no result would break the
- * thread); otherwise they are a fresh user turn. A job asking whether to
- * continue has no call to answer, so it is a turn as well.
- */
-function joinThread(
-  task: { status: TaskStatus; pending: TaskPending | null },
-  answer: string,
-  thread: ModelMessage[],
-): { reply: ModelMessage; as: "answer" | "interruption" | "follow-up" } {
-  const pending = task.status === "waiting" ? task.pending : null;
-  const asks = pending?.toolCallId ? openAsks(thread, pending.toolCallId) : [];
-  if (asks.length) {
-    return {
-      as: "answer",
-      reply: {
-        role: "tool",
-        content: asks.map((toolCallId) => ({
-          type: "tool-result" as const,
-          toolCallId,
-          toolName: TOOL_NAMES.ask_thursday,
-          output: { type: "text" as const, value: answer },
-        })),
+/** One run of a participant's turn from its stored transcript: how it ended, or why it broke. Null when the thread is gone. */
+async function attempt(work: RoomWork, signal: AbortSignal) {
+  const writer = new TranscriptWriter(work, signal);
+  let ending: { text: string; stopped: boolean } | null = null;
+  let failure: { message: string; retry: boolean } | null = null;
+  try {
+    const thread = await findThread(work.threadId);
+    if (!thread) return null;
+    const prior = await listParticipantTranscript(work.threadId, work.bot);
+    if (!prior.length)
+      await appendRoomMessage(work.threadId, {
+        bot: work.bot,
+        parent: work.id,
+        role: "user",
+        content: `Request: ${thread.request}\nCoordinator: ${thread.bot}. Continue as ${work.bot} in this thread.`,
+        hidden: true,
+      });
+    await consumeRoomInbox(work);
+    const history = await listParticipantTranscript(work.threadId, work.bot);
+    await runBot(
+      {
+        bot: work.bot,
+        messages: resumeTranscript(
+          history,
+          await listRoomReceipts(work.threadId, work.bot, history),
+        ),
       },
-    };
+      {
+        signal,
+        threadId: work.threadId,
+        parent: work.id,
+        caller: work.caller,
+        owner: thread.bot,
+        contextBudget: await roomContextBudget(work.threadId, work.bot),
+        session: botBrowserSession(work.threadId, work.bot),
+        notes: () => consumeRoomInbox(work),
+        send: async (input) => {
+          signal.throwIfAborted();
+          let to = input.to.trim();
+          if (to.toLowerCase() === ROOM_THURSDAY.toLowerCase())
+            to = ROOM_THURSDAY;
+          else {
+            const bot = await findJobBot(to);
+            if (!bot || bot.disabled)
+              publicError(`No enabled bot named "${to}".`);
+            to = bot.name;
+          }
+          const receipt = await sendRoomMessage(work, { ...input, to });
+          void pump(work.threadId).catch((cause) =>
+            logger.error("room message scheduling", cause),
+          );
+          return receipt;
+        },
+        emit: async (event) => {
+          if (signal.aborted && event.type !== "tool-result") return;
+          await writer.on(event);
+          if (event.type === "step")
+            await addThreadUsage(
+              work.threadId,
+              event.usage,
+              work.bot === thread.bot
+                ? { tokens: event.usage.input, budget: event.budget }
+                : null,
+            );
+          if (event.type === "compact")
+            await addThreadUsage(work.threadId, event.usage);
+          if (event.type === "turn-end") ending = event;
+          if (event.type === "error") {
+            failure = { message: event.message, retry: event.retry };
+            if (event.budget) await lowerRoomContextBudget(work, event.budget);
+          }
+        },
+      },
+    );
+  } catch (cause) {
+    if (!signal.aborted) {
+      logger.error(`thread ${work.threadId}: ${work.bot} broke`, cause);
+      failure = {
+        message: modelErrorToString(cause),
+        // A public error is the app refusing the turn (no key, no model), not a break
+        retry: !isPublicError(cause) && !isProviderRefusal(cause),
+      };
+    }
   }
   return {
-    as:
-      task.status === "running"
-        ? "interruption"
-        : task.status === "waiting"
-          ? "answer"
-          : "follow-up",
-    reply: { role: "user", content: answer },
+    ending: ending as { text: string; stopped: boolean } | null,
+    failure: failure as { message: string; retry: boolean } | null,
   };
 }
 
-export async function cancelTask(id: string) {
-  return taskLock(id, async () => {
-    const task = await findTask(id);
-    if (!task) publicError("No such job.");
-    // A job that already ended has an answer in `outcome`; cancelling would
-    // write "Cancelled." over the thing it was run for.
-    if (task.endedAt) publicError("That job has already ended.");
-    forgetResume(id);
-    const live = running.get(id);
-    live?.stop.abort();
-    await updateTask(id, {
-      status: "failed",
-      outcome: "Cancelled.",
+/** Streaming updates share stable row slots; tools may be observed before the stream reports them. */
+class TranscriptWriter {
+  private lane = PromiseChain();
+  private assistant: number | null = null;
+  private tool: number | null = null;
+  private parts: {
+    assistant: Exclude<
+      Extract<ModelMessage, { role: "assistant" }>["content"],
+      string
+    >;
+    tool: Extract<ModelMessage, { role: "tool" }>["content"];
+  } = { assistant: [], tool: [] };
+  private calls = new Set<string>();
+  private results = new Set<string>();
+  constructor(
+    private work: RoomWork,
+    private signal: AbortSignal,
+  ) {}
+  private async write(
+    seq: number | null,
+    value: ModelMessage,
+    compact = false,
+  ) {
+    const row = {
+      bot: this.work.bot,
+      parent: this.work.id,
+      role: value.role,
+      content: value.content,
+      compact,
+      note: compact,
+    };
+    if (seq === null) return appendRoomMessage(this.work.threadId, row);
+    await upsertMessage(this.work.threadId, seq, row);
+    return seq;
+  }
+  on(event: ThreadEvent) {
+    return this.lane(async () => {
+      if (this.signal.aborted && event.type !== "tool-result") return;
+      if (event.type === "text") {
+        this.parts.assistant.push({ type: "text", text: event.text });
+        this.assistant = await this.write(this.assistant, {
+          role: "assistant",
+          content: this.parts.assistant,
+        });
+      } else if (event.type === "tool") {
+        if (this.calls.has(event.id)) {
+          const part = this.parts.assistant.find(
+            (part) => part.type === "tool-call" && part.toolCallId === event.id,
+          );
+          if (part && part.type === "tool-call" && event.providerOptions) {
+            part.providerOptions = event.providerOptions;
+            this.assistant = await this.write(this.assistant, {
+              role: "assistant",
+              content: this.parts.assistant,
+            });
+          }
+          return;
+        }
+        this.calls.add(event.id);
+        this.parts.assistant.push({
+          type: "tool-call",
+          toolCallId: event.id,
+          toolName: event.name,
+          input: event.input,
+          ...(event.providerExecuted ? { providerExecuted: true } : {}),
+          ...(event.providerOptions
+            ? { providerOptions: event.providerOptions }
+            : {}),
+        });
+        this.assistant = await this.write(this.assistant, {
+          role: "assistant",
+          content: this.parts.assistant,
+        });
+      } else if (event.type === "tool-result") {
+        if (this.results.has(event.id)) return;
+        this.results.add(event.id);
+        this.parts.tool.push({
+          type: "tool-result",
+          toolCallId: event.id,
+          toolName: event.name,
+          output: event.error
+            ? { type: "error-text", value: String(event.output) }
+            : typeof event.output === "string"
+              ? { type: "text", value: event.output }
+              : { type: "json", value: event.output as never },
+        });
+        this.tool = await this.write(this.tool, {
+          role: "tool",
+          content: this.parts.tool,
+        });
+      } else if (event.type === "step") {
+        const slots = [this.assistant, this.tool];
+        for (const [index, value] of event.messages.entries())
+          await this.write(slots[index] ?? null, value);
+        const stale = slots
+          .slice(event.messages.length)
+          .filter((seq): seq is number => seq !== null);
+        if (stale.length) await deleteMessages(this.work.threadId, stale);
+        this.assistant = this.tool = null;
+        this.parts = { assistant: [], tool: [] };
+      } else if (event.type === "compact") {
+        await this.write(null, { role: "user", content: event.text }, true);
+      }
+    });
+  }
+}
+
+async function stopRuns(id: string) {
+  const live = [...running.values()].filter((run) => run.threadId === id);
+  for (const run of live) run.stop.abort();
+  await Promise.all(live.map((run) => run.done));
+}
+
+export async function cancelThread(id: string) {
+  await threadLock(id, async () => {
+    const thread = await findThread(id);
+    if (!thread) publicError("No such thread.");
+    if (thread.endedAt) publicError("That thread has already ended.");
+    await cancelRoom(id);
+    await updateThread(id, {
+      status: "cancelled",
+      outcome: null,
       pending: null,
-      // Nobody needs to be told what they just did themselves
       seen: true,
       endedAt: new Date(),
     });
-    // Cancel is a real end, so the browser session closes here; the abort in
-    // answerTask is followed by a relaunch and must not close it.
-    await live?.done;
+    await stopRuns(id);
     await closeJobShell(id);
   });
 }
-
-/** Deletes the job, stopping it first if it is still running. */
-export async function removeTask(id: string) {
-  return taskLock(id, () => removeLockedTask(id));
+export async function removeThread(id: string) {
+  return threadLock(id, () => removeLockedThread(id));
 }
-
-/** The caller holds taskLock, so a follow-up cannot race deletion or browser cleanup. */
-async function removeLockedTask(id: string) {
-  const task = await findTask(id);
-  forgetResume(id);
-  const live = running.get(id);
-  live?.stop.abort();
-  await live?.done;
+async function removeLockedThread(id: string) {
+  const thread = await findThread(id);
+  await cancelRoom(id);
+  await stopRuns(id);
   await closeJobShell(id);
-  const gone = await deleteTask(id);
-  if (gone && task) await removeJobScratch(id, task.label);
-  return gone;
+  const removed = await deleteThread(id);
+  if (removed && thread) await removeJobScratch(id, thread.label);
+  return removed;
 }
 
-/**
- * Clearing finished jobs from the screen, and their working folders with them.
- * Lives here rather than in the query because a row and a folder go together
- * and only one of the two is the database's.
- */
-export async function removeFinishedTasks(): Promise<number> {
+export async function removeFinishedThreads(): Promise<number> {
   let removed = 0;
-  for (const task of await listTaskFolders()) {
-    if (task.status !== "done" && task.status !== "failed") continue;
-    await taskLock(task.id, async () => {
-      const current = await findTask(task.id);
-      if (current?.status !== "done" && current?.status !== "failed") return;
-      if (await removeLockedTask(task.id)) removed += 1;
+  for (const thread of await listThreadFolders()) {
+    if (thread.status !== "done" && thread.status !== "cancelled") continue;
+    await threadLock(thread.id, async () => {
+      const current = await findThread(thread.id);
+      if (current?.status !== "done" && current?.status !== "cancelled") return;
+      if (await removeLockedThread(thread.id)) removed += 1;
     });
   }
   return removed;
@@ -343,25 +498,25 @@ export async function removeFinishedTasks(): Promise<number> {
  */
 export async function sweepJobFiles(): Promise<string[]> {
   const cutoff = Date.now() - WORKSPACE_KEEP.forMs;
-  const stale = (task: {
-    status: TaskStatus;
+  const stale = (thread: {
+    status: ThreadStatus;
     endedAt: Date | null;
     updatedAt: Date;
   }) =>
-    (task.status === "done" || task.status === "failed") &&
-    (task.endedAt ?? task.updatedAt).getTime() < cutoff;
+    (thread.status === "done" || thread.status === "cancelled") &&
+    (thread.endedAt ?? thread.updatedAt).getTime() < cutoff;
 
   // Folders on disk; each one a job owns is taken out as its job is read
   const unowned = new Set(await listScratchFolders());
   const removed: string[] = [];
-  for (const task of await listTaskFolders()) {
-    const folder = jobScratch(task.id, task.label);
-    if (!unowned.delete(folder) || !stale(task)) continue;
-    await taskLock(task.id, async () => {
-      const now = await findTask(task.id);
+  for (const thread of await listThreadFolders()) {
+    const folder = jobScratch(thread.id, thread.label);
+    if (!unowned.delete(folder) || !stale(thread)) continue;
+    await threadLock(thread.id, async () => {
+      const now = await findThread(thread.id);
       if (!now || !stale(now)) return;
-      await closeHiddenBrowser(task.id);
-      await removeJobScratch(task.id, task.label);
+      await closeHiddenBrowser(thread.id);
+      await removeJobScratch(thread.id, thread.label);
       removed.push(folder);
     });
   }
@@ -375,600 +530,44 @@ export async function sweepJobFiles(): Promise<string[]> {
   return removed;
 }
 
-/**
- * At boot (instrumentation): rows the last process left as running are not
- * running now. Each is parked the way every stop the app causes is (parkTask),
- * so it picks itself back up once a browser is on the app. Nothing else needs
- * to reconcile — a run that ends badly records it itself (drive), and the write
- * that does so is durable (database/db.ts oneAtATime).
- */
-export async function sweepTasks() {
-  for (const id of await listRunningTaskIds()) {
-    if (!running.has(id)) {
-      await parkTask(id, {
-        why: "The server restarted while this was running.",
-      });
-    }
+/** A server restart is a manual resume boundary: external effects may already have happened. */
+export async function sweepThreads() {
+  for (const id of await listRunningThreadIds()) {
+    await threadLock(id, async () => {
+      if ([...running.values()].some((run) => run.threadId === id)) return;
+      if ((await findThread(id))?.status !== "running") return;
+      await pauseRoom(
+        id,
+        "The server restarted. Resume from the saved conversation.",
+      );
+    });
   }
 }
-
-/**
- * The browser closed, or the server is going down (instrumentation): every run
- * stops where it is and is parked (parkTask). All at once — a run slow to let go
- * must not keep the others running. The thread stays, and so does the window: a
- * resume carries on in the page it left off at, including a sign-in the user is
- * mid-way through, which only that window can finish.
- */
-export async function pauseTasks(reason: string) {
-  const work = Promise.all(
-    [...running].map(([id, run]) =>
-      taskLock(id, async () => {
-        run.stop.abort();
-        await run.done;
-        const task = await findTask(id);
-        if (!task || task.status !== "running") return;
-        await parkTask(id, { why: reason });
-      }).catch((cause) => logger.error(`task ${id}: pause`, cause)),
-    ),
-  ).then(() => {});
-  pausing.current = work;
-  await work;
+export async function pauseThreads(reason: string, auto = false) {
+  const pause = (async () => {
+    const ids = await listRunningThreadIds();
+    await Promise.all(
+      ids.map((id) =>
+        threadLock(id, async () => {
+          if ((await findThread(id))?.status !== "running") return;
+          await stopRuns(id);
+          await pauseRoom(id, reason, auto);
+        }),
+      ),
+    );
+  })();
+  pausing.current = pause;
+  await pause;
 }
-
-/**
- * A browser is on the app again (instrumentation `presence.onBack`): every job
- * the app parked picks itself back up — now, or when the wait a failed model
- * call set runs out. A pause still under way is waited for first, so a quick
- * reload does not miss what it parks.
- */
-export async function resumeStoppedTasks() {
+export async function resumeStoppedThreads() {
   await pausing.current;
-  for (const task of await listAutoStoppedTasks()) {
-    scheduleResume(task.id, task.retryAt - Date.now());
-  }
-}
-
-/** One timer per job: a newer stop replaces what an older one set. */
-function scheduleResume(id: string, inMs: number) {
-  forgetResume(id);
-  resumes.set(
-    id,
-    setTimeout(
-      () => {
-        resumes.delete(id);
-        void resumeTask(id).catch((cause) =>
-          logger.error(`task ${id}: resume`, cause),
-        );
-      },
-      Math.max(0, inMs),
-    ),
-  );
-}
-
-function forgetResume(id: string) {
-  clearTimeout(resumes.get(id));
-  resumes.delete(id);
-}
-
-/**
- * Picks a parked job back up. Only with a browser on the app — nothing runs
- * without one, and its return tries again — and only if it is still parked: a
- * person may have answered, cancelled or deleted it since. No turn is added,
- * because nobody said anything; the note parkTask left is what the run reads last.
- */
-async function resumeTask(id: string) {
   if (!presence.watching) return;
-  await taskLock(id, async () => {
-    const task = await findTask(id);
-    if (!task || task.status !== "waiting" || !task.pending?.auto) return;
-    if (running.has(id)) return;
-    logger.debug(`task ${id}: picking back up`);
-    const thread = await listThread(id);
-    await updateTask(id, {
-      status: "running",
-      outcome: null,
-      pending: null,
-      seen: false,
-      endedAt: null,
-    });
-    launch(id, {
-      bot: task.bot,
-      messages: resumeThread(thread, await listBorrowedTails(id)),
-    });
-  });
-}
-
-/**
- * Stops a job the app stopped, rather than the bot or a person: a restart, a
- * closed browser, a model call a retry or a compaction can fix. It waits with
- * the one continue option like any stop, with a line in its thread saying why —
- * the resumed run reads it — and picks itself back up (scheduleResume). Unless
- * the app has already stopped it `BOT_RUN.autoResumes` times since a person last
- * spoke to it: that is what a job that keeps taking the server down looks like,
- * and it waits for a person instead.
- */
-async function parkTask(
-  id: string,
-  stop: {
-    why: string;
-    /** A failed model call: wait before trying again (config BOT_RUN.retryAfterMs). */
-    retry?: boolean;
-    /** Where the job compacts from now on (bot.run `budget`). */
-    budget?: number;
-  },
-) {
-  const stops = await countStopsSinceSpoken(id);
-  const auto = stops < BOT_RUN.autoResumes;
-  const waits = BOT_RUN.retryAfterMs;
-  const inMs =
-    auto && stop.retry ? waits[Math.min(stops, waits.length - 1)] : 0;
-
-  await writeNote(id, stopNote(stop.why)).catch((cause) =>
-    logger.warn(`task ${id}: stop not written`, cause),
-  );
-  await writeEnding(id, {
-    status: "waiting",
-    outcome: auto
-      ? `${stop.why} ${inMs ? `It tries again in ${spell(inMs)}.` : "It picks back up by itself."}`
-      : `${stop.why} It has stopped ${stops + 1} times in a row, so it waits for you. Everything it had done is still here.`,
-    pending: {
-      toolCallId: null,
-      options: [TASK_CONTINUE],
-      ...(auto ? { auto: true, retryAt: Date.now() + inMs } : {}),
-    },
-    ...(stop.budget ? { contextBudget: stop.budget } : {}),
-    seen: false,
-    endedAt: null,
-  });
-  if (auto) scheduleResume(id, inMs);
-}
-
-const spell = (ms: number) =>
-  ms < 60_000
-    ? `${Math.round(ms / 1000)} seconds`
-    : `${Math.round(ms / 60_000)} minutes`;
-
-/** A line from the app in a thread no run is writing (ThreadWriter.note is the one for a run that is). */
-async function writeNote(id: string, text: string) {
-  await upsertMessage(id, (await lastSeq(id)) + 1, {
-    bot: null,
-    parent: null,
-    role: "user",
-    content: text,
-    note: true,
-  });
-}
-
-/**
- * The write that tells anyone a run ended at all. Tried again: everything else
- * in the run has happened, and a miss leaves the row `running` until the next
- * boot. Writes are serialised, so what is left is a database that briefly
- * cannot be written to (database/db.ts).
- */
-async function writeEnding(
-  id: string,
-  patch: Parameters<typeof updateTask>[1],
-) {
-  for (const wait of [1_000, 3_000, null]) {
-    try {
-      await updateTask(id, patch);
-      return;
-    } catch (cause) {
-      if (wait === null) throw cause;
-      logger.warn(`task ${id}: ending not written, trying again`, cause);
-      await new Promise((resolve) => setTimeout(resolve, wait));
-    }
-  }
-}
-
-function launch(id: string, input: { bot: string; messages: ModelMessage[] }) {
-  const stop = new AbortController();
-  let finish: () => void = () => {};
-  const done = new Promise<void>((resolve) => {
-    finish = resolve;
-  });
-  // Notes said to the job while it runs. The run drains them before each step;
-  // what is left when the loop ends is handed on by `drive`.
-  const notes: string[] = [];
-  const run: Run = {
-    stop,
-    done,
-    note: (text) => {
-      if (!running.has(id)) return false;
-      notes.push(text);
-      return true;
-    },
-  };
-  running.set(id, run);
-  // Held by the server, not by a request: a run also starts from a returning
-  // browser and a retry timer, where `after()` throws (no request) or waits for
-  // the event stream to close
-  void drive(id, input, stop.signal, notes)
-    .catch((cause) => {
-      logger.error(`task ${id}`, cause);
-    })
-    .finally(finish);
-}
-
-/**
- * Writes the thread one step at a time: rows go in while the step streams and
- * are rewritten from the step's own messages when it ends. Slots (`seq`) are
- * claimed synchronously because a borrowed bot writes into the same thread
- * from inside a tool call.
- */
-class ThreadWriter {
-  private seq: number;
-  /** One step in flight per writer: the job's bot and each bot it borrowed. */
-  private steps = new Map<
-    string,
-    { assistant: number | null; tool: number | null; parts: StepParts }
-  >();
-
-  /** Set once the run is over: a borrowed run still unwinding must not write into a thread a resume has taken. */
-  private closed = false;
-
-  constructor(
-    private taskId: string,
-    seq: number,
-  ) {
-    this.seq = seq;
-  }
-
-  close() {
-    this.closed = true;
-  }
-
-  /**
-   * A line from the app rather than from anyone in the room. A user row so a
-   * resumed run reads it as something it was told, and `note` so the screen
-   * draws it as a note and `rosterOf` does not take it for a person speaking.
-   */
-  async note(text: string) {
-    await upsertMessage(this.taskId, ++this.seq, {
-      bot: null,
-      parent: null,
-      role: "user",
-      content: text,
-      note: true,
+  for (const id of await listAutoStoppedThreads()) {
+    await threadLock(id, async () => {
+      const current = await findThread(id);
+      if (current?.status !== "waiting" || !current.pending?.auto) return;
+      await resumeRoom(id, false);
     });
+    await pump(id);
   }
-
-  private step(event: TaskEvent) {
-    const key = `${event.bot} ${event.parent ?? ""}`;
-    let step = this.steps.get(key);
-    if (!step) {
-      step = {
-        assistant: null,
-        tool: null,
-        parts: { assistant: [], tool: [] },
-      };
-      this.steps.set(key, step);
-    }
-    return step;
-  }
-
-  private write(
-    seq: number,
-    event: TaskEvent,
-    message: ModelMessage,
-    extra: { compact?: boolean; note?: boolean } = {},
-  ) {
-    const input: TaskMessageInput = {
-      bot: event.bot,
-      parent: event.parent,
-      role: message.role,
-      content: message.content,
-      compact: extra.compact ?? false,
-      note: extra.note ?? false,
-    };
-    return upsertMessage(this.taskId, seq, input);
-  }
-
-  async on(event: TaskEvent) {
-    if (this.closed) return;
-    const step = this.step(event);
-    switch (event.type) {
-      case "request":
-        return this.write(++this.seq, event, {
-          role: "user",
-          content: event.content,
-        });
-
-      case "text":
-        step.assistant ??= ++this.seq;
-        step.parts.assistant.push({ type: "text", text: event.text });
-        return this.write(step.assistant, event, {
-          role: "assistant",
-          content: step.parts.assistant,
-        });
-
-      case "tool":
-        step.assistant ??= ++this.seq;
-        step.parts.assistant.push({
-          type: "tool-call",
-          toolCallId: event.id,
-          toolName: event.name,
-          input: event.input,
-        });
-        return this.write(step.assistant, event, {
-          role: "assistant",
-          content: step.parts.assistant,
-        });
-
-      case "tool-result":
-        step.assistant ??= ++this.seq;
-        step.tool ??= ++this.seq;
-        step.parts.tool.push({
-          type: "tool-result",
-          toolCallId: event.id,
-          toolName: event.name,
-          output: event.error
-            ? { type: "error-text", value: String(event.output) }
-            : typeof event.output === "string"
-              ? { type: "text", value: event.output }
-              : { type: "json", value: event.output as never },
-        });
-        return this.write(step.tool, event, {
-          role: "tool",
-          content: step.parts.tool,
-        });
-
-      case "step": {
-        // The slots the step streamed into, then fresh ones for anything more
-        const slots = [step.assistant, step.tool];
-        this.steps.delete(`${event.bot} ${event.parent ?? ""}`);
-        const written = new Set<number>();
-        for (const [index, message] of event.messages.entries()) {
-          const seq = slots[index] ?? ++this.seq;
-          written.add(seq);
-          await this.write(seq, event, message);
-        }
-        // A slot claimed while streaming that the step did not fill: a
-        // provider-executed tool (web search) answers inside the assistant
-        // message, leaving a tool row with no call in front of it.
-        const stale = slots.filter(
-          (seq): seq is number => seq !== null && !written.has(seq),
-        );
-        if (stale.length) await deleteMessages(this.taskId, stale);
-        return;
-      }
-
-      case "compact":
-        // The row a resume restarts from (task.query listThread); earlier rows
-        // stay for the screen.
-        return this.write(
-          ++this.seq,
-          event,
-          { role: "user", content: event.text },
-          { compact: true, note: true },
-        );
-
-      case "interjection":
-        // Between the two steps, where the model read it.
-        return this.write(++this.seq, event, {
-          role: "user",
-          content: event.text,
-        });
-
-      default:
-        return;
-    }
-  }
-}
-
-type StepParts = {
-  assistant: Extract<ModelMessage, { role: "assistant" }>["content"] &
-    unknown[];
-  tool: Extract<ModelMessage, { role: "tool" }>["content"];
-};
-
-async function drive(
-  id: string,
-  input: { bot: string; messages: ModelMessage[] },
-  signal: AbortSignal,
-  notes: string[],
-) {
-  const thread = new ThreadWriter(id, await lastSeq(id));
-  const desks = createBotDesks();
-
-  let ending: {
-    status: TaskStatus;
-    outcome: string;
-    pending: TaskPending | null;
-  } | null = null;
-  /**
-   * What broke, as the provider said it — the thread carries it too (below) —
-   * and whose it is to fix (model.ts modelFailureOf).
-   */
-  let broken: {
-    message: string;
-    failure: ModelFailure;
-    budget?: number;
-  } | null = null;
-
-  try {
-    await runBot(input, {
-      signal,
-      taskId: id,
-      desks,
-      notes: () => notes.splice(0),
-      emit: async (event) => {
-        await thread.on(event);
-        // A borrowed bot's steps count toward the job's usage; the window fill
-        // is only the job's own bot's.
-        if (event.type === "step") {
-          await addTaskUsage(
-            id,
-            event.usage,
-            event.parent
-              ? null
-              : { tokens: event.usage.input, budget: event.budget },
-          );
-        }
-        // Compaction and answering a borrowed bot (bot.run answerBack) cost the job too.
-        if (event.type === "compact" || event.type === "answered") {
-          await addTaskUsage(id, event.usage);
-        }
-        // Only the job's own bot ends it; a borrowed bot's ending is a tool result.
-        if (event.parent) return;
-        if (event.type === "waiting") {
-          ending = {
-            status: "waiting",
-            outcome: event.question,
-            pending: { toolCallId: event.id, options: event.options },
-          };
-        } else if (event.type === "answer") {
-          // An answer ends the job; one the app stopped (bot.run `stopped`) waits
-          // with the continue option. The outcome is the answer text verbatim;
-          // the same text is the thread line (bot.query linesOf).
-          ending = event.stopped
-            ? {
-                status: "waiting",
-                outcome: event.text,
-                pending: { toolCallId: null, options: [TASK_CONTINUE] },
-              }
-            : { status: "done", outcome: event.text, pending: null };
-        } else if (event.type === "error") {
-          broken = {
-            message: event.message,
-            failure: event.failure,
-            budget: event.budget,
-          };
-          ending = {
-            status: "failed",
-            outcome: broke(event.message),
-            pending: null,
-          };
-        }
-      },
-    });
-  } catch (cause) {
-    broken = {
-      message: modelErrorToString(cause),
-      failure: modelFailureOf(cause),
-    };
-    ending = {
-      status: "failed",
-      outcome: broke(broken.message),
-      pending: null,
-    };
-  } finally {
-    // A resumed segment cannot overtake a participant still unwinding its tools.
-    await desks.settle();
-    thread.close();
-    running.delete(id);
-  }
-
-  // Whoever aborted writes the row: cancel marks failed, answerTask relaunches,
-  // a pause parks.
-  if (signal.aborted) return;
-
-  const final = ending ?? {
-    status: "failed" as const,
-    outcome: "Stopped without handing anything back.",
-    pending: null,
-  };
-  // A break a moment or a compaction can fix is the app's to pick back up, not
-  // a failure for a person to find (parkTask)
-  const park = broken && broken.failure !== "fatal" ? broken : null;
-
-  try {
-    if (park) {
-      await parkTask(id, {
-        why:
-          park.failure === "overflow"
-            ? `The model refused the context as too long (${park.message}); it is compacted before going on.`
-            : `The model call failed (${park.message}).`,
-        retry: park.failure === "transient",
-        budget: park.budget,
-      });
-    } else {
-      // A thread that simply stops cannot be read: the room shows a tool call
-      // with no answer, and a resumed run has no idea why. The task row says it
-      // too, but the row is a status and this is the story. Never at the cost of
-      // the status write, which is what tells anyone the job ended at all.
-      if (final.status === "failed") {
-        await thread
-          .note(`The run stopped here: ${broken?.message ?? final.outcome}`)
-          .catch((cause) =>
-            logger.warn(`task ${id}: break not written`, cause),
-          );
-      }
-      await writeEnding(id, {
-        ...final,
-        seen: false,
-        endedAt: final.status === "waiting" ? null : new Date(),
-      });
-    }
-  } catch (cause) {
-    // Said plainly rather than thrown into a promise nobody awaits; the row
-    // stays `running`, which the next boot reconciles (sweepTasks)
-    logger.error(`task ${id}: could not record how it ended`, cause);
-    return;
-  }
-
-  // Follow-ups can reopen a completed job. Its participants keep their browsers
-  // until cancellation, deletion, or the workspace retention sweep.
-  void pruneJobFiles();
-
-  const task = await findTask(id);
-  if (!task) return;
-
-  // Notes that arrived after the last step began were never read; they go on
-  // as a turn and the job resumes from what it just handed back.
-  if (notes.length) {
-    const said = notes.splice(0).join("\n\n");
-    logger.debug(`task ${id}: interjection after the run ended — resuming`);
-    void answerTask(id, said).catch((cause) => {
-      logger.error(`task ${id}: late interjection`, cause);
-    });
-    return;
-  }
-
-  // A parked job picks itself back up; nobody needs telling yet
-  if (park) return;
-
-  await announce(task.label, final);
-
-  // A finished job naming a document opens it on the screen (workspace artifact-view).
-  if (final.status === "done") {
-    const path = artifactIn(final.outcome);
-    if (path) {
-      appEvents.emit({ type: "artifact", taskId: id, label: task.label, path });
-    }
-  }
-}
-
-/** The document an answer names: the artifacts folder first, otherwise the first drawable file. */
-function artifactIn(outcome: string): string | null {
-  const paths = pathsIn(outcome);
-  return (
-    paths.find((path) => path.startsWith(`${PATHS.artifacts}/`)) ??
-    paths[0] ??
-    null
-  );
-}
-
-/**
- * A failure the user can act on: the provider message stays as the clue, plus
- * the way out (answering resumes from a condensed thread).
- */
-function broke(message: string): string {
-  return `The run broke: ${message}\n\nEverything it had done is still here — answer to pick it back up.`;
-}
-
-/**
- * Desktop notification unless a call is live (the browser relays the job to
- * the call itself). `watchers === 0` is checked first: a crashed browser can
- * leave a live call row behind until boot clears it (thursday.query sweepCalls).
- */
-async function announce(
-  label: string,
-  ending: { status: TaskStatus; outcome: string },
-) {
-  const nobodyWatching = appEventStream.watchers === 0;
-  if (!nobodyWatching && (await isAnyCallLive())) return;
-
-  const title = ending.status === "waiting" ? `${label} — needs you` : label;
-  desktopNotify(title, ending.outcome);
 }